Model Context Protocol + .faf

We are the 'C' in MCP • Context, authored.

The official MCP documentation for .faf — persistent project context across Claude, Grok, Gemini, Cursor, Windsurf, Cline, VS Code, and every MCP-compatible platform.

IANA application/vnd.faf+yaml MCP Registry merged #2759

What is MCP + .faf?

MCP is an open protocol connecting AI assistants to context. .faf is the IANA-registered context format (application/vnd.faf+yaml) — one source, every model and tool.

.faf owns the write side. The Context-Ingestion contract owns the read side. faf-cli authors it — Claude-first, cross-vendor.

The MCP landscape

Two questions place any MCP server: what it's built with, and how it runs. .faf is the context underneath — the same file whichever answers you pick.

Built with

  • Official SDKs — TypeScript, Python, C#, Go and Rust are Tier 1; Java, Ruby, Swift, PHP and Kotlin follow. Same protocol, each in its own language's idiom.
  • FastMCP — the high-level Python framework (Prefect-maintained, now on v4). It created the API that was folded into the official Python SDK in 2024, and carries the fuller feature line — most Python MCP servers use it.
  • Hand-rolled — a server can speak the wire protocol directly. Rust and Go servers often do.

Runs as

  • Local — stdio — the client launches the server as a child process (npx, uvx). Nothing to host. The common case.
  • Remote — Streamable HTTP — a hosted endpoint (Cloudflare Workers, Vercel, your own). Protocol 2026-07-28 is sessionless. The older SSE transport is deprecated.

Where the FAF family sits: claude-faf-mcp, faf-mcp and grok-faf-mcp on the TypeScript SDK · gemini-faf-mcp on FastMCP · rust-faf-mcp hand-rolled in Rust. Every one reads the same .faf.

Deeper: modelcontextprotocol.io · the SDKs · gofastmcp.com

The Trinity: Model → Context → Protocol

Model / Tool
Context
Protocol
Models
Claude
.faf
MCP
Grok
.faf
MCP
Gemini
.faf
MCP
Tools & IDEs
Cursor
.faf
MCP
Windsurf
.faf
MCP
Cline
.faf
MCP
VS Code
.faf
MCP
Any LLM
IANA Format
Open Protocol

.faf is the Context layer for any Model using the MCP Protocol

Formats = facts.

The MCP Family

One IANA-registered format. A server for every model — plus the CLI that authors it.

📦

claude-faf-mcp Claude

The Claude-first MCP server — persistent .faf context for Claude Desktop and Claude Code.

  • ☑️ Claude-first — Desktop + Claude Code
  • ☑️ In Anthropic's MCP Registry — merged #2759
  • ☑️ bi-sync .faf ↔ CLAUDE.md
npx claude-faf-mcp
npm → GitHub →
📦

faf-mcp IDEs

The MCP server for Cursor, Windsurf, Cline, VS Code, and any MCP-compatible client.

  • ☑️ Cursor, Windsurf, Cline, VS Code — any MCP client
  • ☑️ bi-sync .faf ↔ your context files
  • ☑️ In the official MCP Registry
npx faf-mcp
npm → GitHub →
📦

grok-faf-mcp Grok

The MCP server for xAI Grok — the same .faf context, read by Grok.

  • ☑️ Same IANA .faf format
  • ☑️ Cross-vendor context, one file
  • ☑️ In the official MCP Registry
npx grok-faf-mcp
npm → GitHub →
📦

gemini-faf-mcp Gemini

The MCP server for Google Gemini — Python, published to PyPI.

  • ☑️ Python — uvx gemini-faf-mcp (PyPI)
  • ☑️ Reference implementation of the Zenodo Context paper
  • ☑️ Companion: Agents paper (.fafa)
  • ☑️ Same IANA .faf format
uvx gemini-faf-mcp
PyPI → GitHub →
📦

WJTTC Testing

Championship-grade MCP testing. If a server passes WJTTC, it works.

  • ☑️ GitHub Action included
  • ☑️ Verifies MCP servers actually work
  • ☑️ In the official MCP Registry
npx wjttc
npm → GitHub →
📦

faf-cli Terminal

The CLI on-ramp — author, score, and sync your .faf from any terminal. Claude-first, works cross-vendor.

  • ☑️ The on-ramp — faf init, faf score, faf sync
  • ☑️ bi-sync .faf ↔ CLAUDE.md
  • ☑️ Free forever · open source
npm install -g faf-cli
brew install wolfe-jam/faf/faf-cli
npm → GitHub →

Also in the family: rust-faf-mcp (Rust · crates.io) — six FAF servers total in the official MCP Registry.

Tools: core ~12–14 per server, family ~29–33 total. Each server's README is the source of truth.

Installation by Platform

Cursor IDE Setup

1

Install faf-mcp

npm install -g faf-mcp
2

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "faf": {
      "command": "npx",
      "args": ["-y", "faf-mcp"]
    }
  }
}
3

Restart Cursor

Then start prompting: "Use FAF to initialize your project"

Cursor MCP Documentation →

The "Use FAF" Pattern

Just start prompts with Use FAF.

Example commands

  • Use FAF to initialize your project
  • Use FAF to sync my project.faf to all platforms
  • Use FAF to score my AI-readiness
  • Use FAF to enhance my context

Further reading

Get Started

Install faf-mcp → GitHub →